4CPS-149 External validation of population pharmacokinetic models of high dosing methotrexate in paediatric patients with acute lymphoblastic leukaemia

<h3>Background and Importance</h3> High-dose methotrexate (HDMTX) in a 24-hour infusion is essential in the treatment of acute lymphoblastic leukaemia (ALL). To prevent toxicities monitoring methotrexate plasma concentrations (PCmtx) must be performed. <h3>Aim and Objectives</h3> To assess the predictive ability of two methotrexate pharmacokinetic models in paediatric oncology patients at a third-level hospital, utilising an external dataset in the context of our current clinical practice. <h3>Material and Methods</h3> Two HDMTX pharmacokinetics models (Model A and Model B) used in our current clinical practice were selected. Methotrexate plasma concentration (PCmtx) and creatinine levels were analysed 24–48h post-infusion, with high-risk patients assessed starting at 2h. We estimated individual predicted concentrations (Cipred) for each model using Bayesian estimation, then calculated the individual prediction error by assessing the difference between Cipred and the observed concentrations. The prediction’s accuracy and precision were evaluated using the median of individual prediction error (MDIPE) and the absolute median of individual prediction error (MAIPE) respectively. Predictive performance of a model was considered as satisfied when MDIPE%&lt;±10% and MAIPE%&lt;25%. <h3>Results</h3> A total of 560 PCmtx from 57 patients (60% male, aged 2–17) with ALL who received HDMTX (1–5g/m2) were included. Model A demonstrated accuracy and precision over time, with an MDIPE of 2,4% (95% CI: 0,45–4,34) and a MAIPE of 8% showing statistical significance. Model A’s performance consistently met criteria as MDIPE 24h at 6,7% (95% CI: 3,767–9,633), MAIPE 24h at 8,7%, MDIPE 42h at –2,2% (95% CI: −8,608–4,208) and MAIPE 42h at 7,5% were under the limit. The non-significant MDIPE 42h result showed that concentration variances may be due to random variations. Model B, on the other hand, did not meet criteria at 24h (MDIPE 24h 31,1% (95% CI: 22,139–40,061) and MAIPE 24h 31,1%) but improved results at 42h (MDIPE 42h at 6,3% (95% CI: −0,771–13,371) and MAIPE 42h at 9,7%. These results indicate that Model A’s accuracy and precision are both higher than Model B’s although Model B improved its performance over time. <h3>Conclusion and Relevance</h3> Model A’s predictive ability is higher than Model B’s at all times demonstrated by the results. Model A exhibited high accuracy and precision, becoming a superior and more reliable predictive model. <h3>References and/or Acknowledgements</h3> <h3>Conflict of Interest</h3> No conflict of interest